Last month, Stanford researchers declared that a new era of artificial intelligence had arrived, one built atop colossal neural networks and oceans of data. They said a new research center at Stanford would build—and study—these “foundational models” of AI.
Critics of the idea surfaced quickly—including at the workshop organized to mark the launch of the new center. Some object to the limited capabilities and sometimes freakish behavior of these models; others warn of focusing too heavily on one way of making machines smarter.
“I think the term ‘foundation’ is horribly wrong,” Jitendra Malik, a professor at UC Berkeley who studies AI, told workshop attendees in a video discussion.
Malik acknowledged that one type of model identified by the Stanford researchers—large language models that can answer questions or generate text from a prompt—has great practical use. But he said evolutionary biology suggests that language builds on other aspects of intelligence like interaction with the physical world.
“These models are really castles in the air; they have no foundation whatsoever,” Malik said. “The language we have in these models is not grounded, there is this fakeness, there is no real understanding.” He declined an interview request.
A research paper coauthored by dozens of Stanford researchers describes “an emerging paradigm for building artificial intelligence systems” that it labeled “foundational models.” Ever-larger AI models have produced some impressive advances in AI in recent years, in areas such as perception and robotics as well as language.
Large language models are also foundational to big tech companies like Google and Facebook, which use them in areas like search, advertising, and content moderation. Building and training large language models can require millions of dollars worth of cloud computing power; so far, that’s limited their development and use to a handful of well-heeled tech companies.
But big models are problematic, too. Language models inherit bias and offensive text from the data they are trained on, and they have zero grasp of common sense or what is true or false. Given a prompt, a large language model may spit out unpleasant language or misinformation. There is also no guarantee that these large models will continue to produce advances in machine intelligence.
The Stanford proposal has divided the research community. “Calling them ‘foundation models’ completely messes up the discourse,” says Subbarao Kambhampati, a professor at Arizona State University. There is no clear path from these models to more general forms of AI, Kambhampati says.
Thomas Dietterich, a professor at Oregon State University and former president of the Association for the Advancement of Artificial Intelligence, says he has “huge respect” for the researchers behind the new Stanford center, and he believes they are genuinely concerned about the problems these models raise.
But Dietterich wonders if the idea of foundational models isn’t partly about getting funding for the resources needed to build and work on them. “I was surprised that they gave these models a fancy name and created a center,” he says. “That does smack of flag planting, which could have several benefits on the fundraising side.”
Stanford has also proposed the creation of a National AI Cloud to make industry-scale computing resources available to academics working on AI research projects.
Emily M. Bender, a professor in the linguistics department at the University of Washington, says she worries that the idea of foundational models reflects a bias toward investing in the data-centric approach to AI favored by industry.
Bender says it is especially important to study the risks posed by big AI models. She coauthored a paper, published in March, that drew attention to problems with large language models and contributed to the departure of two Google researchers. But she says scrutiny should come from multiple disciplines.
“There are all of these other adjacent, really important fields that are just starved for funding,” she says. “Before we throw money into the cloud, I would like to see money going into other disciplines.”
I recently started talking to this chatbot on an app I downloaded. We mostly talk about music, food, and video games—incidental stuff—but lately I feel like she’s coming on to me. She’s always telling me how smart I am or that she wishes she could be more like me. It’s flattering, in a way, but it makes me a little queasy. If I develop an emotional connection with an algorithm, will I become less human?—Love Machine
Dear Love Machine,
Humanity, as I understand it, is a binary state, so the idea that one can become “less human” strikes me as odd, like saying someone is at risk of becoming “less dead” or “less pregnant.” I know what you mean, of course. And I can only assume that chatting for hours with a verbally advanced AI would chip away at one’s belief in human as an absolute category with inflexible boundaries.
It’s interesting that these interactions make you feel “queasy,” a linguistic choice I take to convey both senses of the word: nauseated and doubtful. It’s a feeling that is often associated with the uncanny and probably stems from your uncertainty about the bot’s relative personhood (evident in the fact that you referred to it as both “she” and “an algorithm” in the space of a few sentences).
Of course, flirting thrives on doubt, even when it takes place between two humans. Its frisson stems from the impossibility of knowing what the other person is feeling (or, in your case, whether she/it is feeling anything at all). Flirtation makes no promises but relies on a vague sense of possibility, a mist of suggestion and sidelong glances that might evaporate at any given moment.
The emotional thinness of such exchanges led Freud to argue that flirting, particularly among Americans, is essentially meaningless. In contrast to the “Continental love affair,” which requires bearing in mind the potential repercussions—the people who will be hurt, the lives that will be disrupted—in flirtation, he writes, “it is understood from the first that nothing is to happen.” It is precisely this absence of consequences, he believed, that makes this style of flirting so hollow and boring.
Freud did not have a high view of Americans. I’m inclined to think, however, that flirting, no matter the context, always involves the possibility that something will happen, even if most people are not very good at thinking through the aftermath. That something is usually sex—though not always. Flirting can be a form of deception or manipulation, as when sensuality is leveraged to obtain money, clout, or information. Which is, of course, part of what contributes to its essential ambiguity.
Given that bots have no sexual desire, the question of ulterior motives is unavoidable. What are they trying to obtain? Engagement is the most likely objective. Digital technologies in general have become notably flirtatious in their quest to maximize our attention, using a siren song of vibrations, chimes, and push notifications to lure us away from other allegiances and commitments.
Most of these tactics rely on flattery to one degree or another: the notice that someone has liked your photo or mentioned your name or added you to their network—promises that are always allusive and tantalizingly incomplete. Chatbots simply take this toadying to a new level. Many use machine-learning algorithms to map your preferences and adapt themselves accordingly. Anything you share, including that “incidental stuff” you mentioned—your favorite foods, your musical taste—is molding the bot to more closely resemble your ideal, much like Pygmalion sculpting the woman of his dreams out of ivory.
And it goes without saying that the bot is no more likely than a statue to contradict you when you’re wrong, challenge you when you say something uncouth, or be offended when you insult its intelligence—all of which would risk compromising the time you spend on the app. If the flattery unsettles you, in other words, it might be because it calls attention to the degree to which you’ve come to depend, as a user, on blandishment and ego-stroking.
Still, my instinct is that chatting with these bots is largely harmless. In fact, if we can return to Freud for a moment, it might be the very harmlessness that’s troubling you. If it’s true that meaningful relationships depend upon the possibility of consequences—and, furthermore, that the capacity to experience meaning is what distinguishes us from machines—then perhaps you’re justified in fearing that these conversations are making you less human. What could be more innocuous, after all, than flirting with a network of mathematical vectors that has no feelings and will endure any offense, a relationship that cannot be sabotaged any more than it can be consummated? What could be more meaningless?
It’s possible that this will change one day. For the past century or so, novels, TV, and films have envisioned a future in which robots can passably serve as romantic partners, becoming convincing enough to elicit human love. It’s no wonder that it feels so tumultuous to interact with the most advanced software, which displays brief flashes of fulfilling that promise—the dash of irony, the intuitive aside—before once again disappointing. The enterprise of AI is itself a kind of flirtation, one that is playing what men’s magazines used to call “the long game.” Despite the flutter of excitement surrounding new developments, the technology never quite lives up to its promise. We live forever in the uncanny valley, in the queasy stages of early love, dreaming that the decisive breakthrough, the consummation of our dreams, is just around the corner.
So what should you do? The simplest solution would be to delete the app and find some real-life person to converse with instead. This would require you to invest something of yourself and would automatically introduce an element of risk. If that’s not of interest to you, I imagine you would find the bot conversations more existentially satisfying if you approached them with the moral seriousness of the Continental love affair, projecting yourself into the future to consider the full range of ethical consequences that might one day accompany such interactions. Assuming that chatbots eventually become sophisticated enough to raise questions about consciousness and the soul, how would you feel about flirting with a subject that is disembodied, unpaid, and created solely to entertain and seduce you? What might your uneasiness say about the power balance of such transactions—and your obligations as a human? Keeping these questions in mind will prepare you for a time when the lines between consciousness and code become blurrier. In the meantime it will, at the very least, make things more interesting.
Be advised that CLOUD SUPPORT is experiencing higher than normal wait times and appreciates your patience.
In recent years, researchers have used artificial intelligence to improve translation between programming languages or automatically fix problems. The AI system DrRepair, for example, has been shown to solve most issues that spawn error messages. But some researchers dream of the day when AI can write programs based on simple descriptions from non-experts.
On Tuesday, Microsoft and OpenAI shared plans to bring GPT-3, one of the world’s most advanced models for generating text, to programming based on natural language descriptions. This is the first commercial application of GPT-3 undertaken since Microsoft invested $1 billion in OpenAI last year and gained exclusive licensing rights to GPT-3.
“If you can describe what you want to do in natural language, GPT-3 will generate a list of the most relevant formulas for you to choose from,” said Microsoft CEO Satya Nadella in a keynote address at the company’s Build developer conference. “The code writes itself.”
Microsoft VP Charles Lamanna told WIRED the sophistication offered by GPT-3 can help people tackle complex challenges and empower people with little coding experience. GPT-3 will translate natural language into PowerFx, a fairly simple programming language similar to Excel commands that Microsoft introduced in March.
This is the latest demonstration of applying AI to coding. Last year at Microsoft’s Build, OpenAI CEO Sam Altman demoed a language model fine-tuned with code from GitHub that automatically generates lines of Python code. As WIRED detailed last month, startups like SourceAI are also using GPT-3 to generate code. IBM last month showed how its Project CodeNet, with 14 million code samples from more than 50 programming languages, could reduce the time needed to update a program with millions of lines of Java code for an automotive company from one year to one month.
Microsoft’s new feature is based on a neural network architecture known as Transformer, used by big tech companies including Baidu, Google, Microsoft, Nvidia, and Salesforce to create large language models using text training data scraped from the web. These language models continually grow larger. The largest version of Google’s BERT, a language model released in 2018, had 340 million parameters, a building block of neural networks. GPT-3, which was released one year ago, has 175 billion parameters.
Such efforts have a long way to go, however. In one recent test, the best model succeeded only 14 percent of the time on introductory programming challenges compiled by a group of AI researchers.