superintelligent AI

Will superintelligent AI sneak up on us? New study offers reassurance

Posted by

0


Improvements in the performance of large language models such as ChatGPT are more predictable than they seem.

A happy red robot goes a different direction against lines of many identical robots.
Some researchers think that AI could eventually achieve general intelligence, matching and even exceeding humans on most tasks.Credit: Charles Taylor/Alamy

Will an artificial intelligence (AI) superintelligence appear suddenly, or will scientists see it coming, and have a chance to warn the world? That’s a question that has received a lot of attention recently, with the rise of large language models, such as that behind ChatGPT, which have achieved vast new abilities as their size has grown. Some findings point to ‘emergence’, a phenomenon in which AI models gain intelligence in a sharp and unpredictable way. But a recent study calls these cases mirages — artefacts arising from how the systems are tested — and suggests that innovative abilities instead build more gradually.

“I think they did a good job of saying ‘nothing magical has happened’,” says Deborah Raji, a computer scientist at the Mozilla Foundation in San Francisco, California, who studies the auditing of artificial intelligence. It’s “a really good, solid, measurement-based critique”.

The work was presented last week at the NeurIPS 2023 machine-learning conference in New Orleans, Louisiana.

Bigger is better

Large language models are typically trained using huge amounts of text, or other information, which they use to generate realistic answers by predicting what comes next. Even without explicit training, they manage to translate language, solve mathematical problems and write poetry or computer code. The bigger the model is — some have more than one hundred billion tunable parameters — the better it performs. Some researchers suspect that these tools will eventually achieve artificial general intelligence (AGI), matching and even exceeding humans on most tasks.ChatGPT broke the Turing test — the race is on for new ways to assess AI

The new research tested claims of emergence in several ways. In one approach, the scientists compared the abilities of four sizes of the GPT-3 model, developed by OpenAI in San Francisco, to add up four-digit numbers. Looking at absolute accuracy, performance differed between the third and fourth size of the model from nearly 0% to nearly 100%. But this trend was less extreme if the number of correctly predicted digits in the answer was considered instead. The researchers also found that they could dampen the curve by giving the models many more test questions — in this case, the smaller models answered correctly some of the time.

Next, the researchers looked at the performance of Google’s LaMDA language model on several tasks. The ones for which it showed a sudden jump in apparent intelligence, such as detecting irony or translating proverbs, were often multiple-choice tasks, with answers scored discretely as right or wrong. When, instead, the researchers examined the probabilities that the models placed on each answer — a continuous metric — signs of emergence disappeared.

Finally, the researchers turned to computer vision, a field in which there are fewer claims of emergence. They trained models to compress and then reconstruct images. By merely setting a strict threshold for correctness, they could induce apparent emergence. “They were creative in the way that they designed their investigation,” says Yejin Choi, a computer scientist at the University of Washington in Seattle who studies AI and common sense.

Nothing ruled out

Study co-author Sanmi Koyejo, a computer scientist at Stanford University in Palo Alto, California, says that it wasn’t unreasonable for people to accept the idea of emergence, given that some systems exhibit abrupt “phase changes”. He also notes that the study can’t completely rule out emergence in large language models — let alone in future systems — but adds that “scientific study to date strongly suggests most aspects of language models are indeed predictable”.

Raji is happy to see the AI community pay more attention to benchmarking, rather than to developing neural-network architectures. She’d like researchers to go even further and ask how well the tasks relate to real-world deployment. For example, does acing the LSAT exam for aspiring lawyers, as GPT-4 has done, mean that a model can act as a paralegal?

The work also has implications for AI safety and policy. “The AGI crowd has been leveraging the emerging-capabilities claim,” Raji says. Unwarranted fear could lead to stifling regulations or divert attention from more pressing risks. “The models are making improvements, and those improvements are useful,” she says. “But they’re not approaching consciousness yet.”

Silicon Valley Is Turning Into Its Own Worst Fear

Posted by

0


We asked a group of writers to consider the forces that have shaped our lives in 2017. Here, science fiction writer Ted Chiang looks at capitalism, Silicon Valley, and its fear of superintelligent AI.

Justin Metz for BuzzFeed News

This summer, Elon Musk spoke to the National Governors Association and told them that “AI is a fundamental risk to the existence of human civilization.” Doomsayers have been issuing similar warnings for some time, but never before have they commanded so much visibility. Musk isn’t necessarily worried about the rise of a malicious computer like Skynet from The Terminator. Speaking to Maureen Dowd for a Vanity Fair article published in April, Musk gave an example of an artificial intelligence that’s given the task of picking strawberries. It seems harmless enough, but as the AI redesigns itself to be more effective, it might decide that the best way to maximize its output would be to destroy civilization and convert the entire surface of the Earth into strawberry fields. Thus, in its pursuit of a seemingly innocuous goal, an AI could bring about the extinction of humanity purely as an unintended side effect.

When Silicon Valley tries to imagine superintelligence, what it comes up with is no-holds-barred capitalism.

This scenario sounds absurd to most people, yet there are a surprising number of technologists who think it illustrates a real danger. Why? Perhaps it’s because they’re already accustomed to entities that operate this way: Silicon Valley tech companies.

Consider: Who pursues their goals with monomaniacal focus, oblivious to the possibility of negative consequences? Who adopts a scorched-earth approach to increasing market share? This hypothetical strawberry-picking AI does what every tech startup wishes it could do — grows at an exponential rate and destroys its competitors until it’s achieved an absolute monopoly. The idea of superintelligence is such a poorly defined notion that one could envision it taking almost any form with equal justification: a benevolent genie that solves all the world’s problems, or a mathematician that spends all its time proving theorems so abstract that humans can’t even understand them. But when Silicon Valley tries to imagine superintelligence, what it comes up with is no-holds-barred capitalism.

In psychology, the term “insight” is used to describe a recognition of one’s own condition, such as when a person with mental illness is aware of their illness. More broadly, it describes the ability to recognize patterns in one’s own behavior. It’s an example of metacognition, or thinking about one’s own thinking, and it’s something most humans are capable of but animals are not. And I believe the best test of whether an AI is really engaging in human-level cognition would be for it to demonstrate insight of this kind.

Insight is precisely what Musk’s strawberry-picking AI lacks, as do all the other AIs that destroy humanity in similar doomsday scenarios. I used to find it odd that these hypothetical AIs were supposed to be smart enough to solve problems that no human could, yet they were incapable of doing something most every adult has done: taking a step back and asking whether their current course of action is really a good idea. Then I realized that we are already surrounded by machines that demonstrate a complete lack of insight, we just call them corporations. Corporations don’t operate autonomously, of course, and the humans in charge of them are presumably capable of insight, but capitalism doesn’t reward them for using it. On the contrary, capitalism actively erodes this capacity in people by demanding that they replace their own judgment of what “good” means with “whatever the market decides.”

It’s assumed that the AI’s approach will be “the question isn’t who is going to let me, it’s who is going to stop me,” i.e., the mantra of Ayn Randian libertarianism that is so popular in Silicon Valley.

Because corporations lack insight, we expect the government to provide oversight in the form of regulation, but the internet is almost entirely unregulated. Back in 1996, John Perry Barlow published a manifesto saying that the government had no jurisdiction over cyberspace, and in the intervening two decades that notion has served as an axiom to people working in technology. Which leads to another similarity between these civilization-destroying AIs and Silicon Valley tech companies: the lack of external controls. If you suggest to an AI prognosticator that humans would never grant an AI so much autonomy, the response will be that you fundamentally misunderstand the situation, that the idea of an ‘off’ button doesn’t even apply. It’s assumed that the AI’s approach will be “the question isn’t who is going to let me, it’s who is going to stop me,” i.e., the mantra of Ayn Randian libertarianism that is so popular in Silicon Valley.

The ethos of startup culture could serve as a blueprint for civilization-destroying AIs. “Move fast and break things” was once Facebook’s motto; they later changed it to “Move fast with stable infrastructure,” but they were talking about preserving what they had built, not what anyone else had. This attitude of treating the rest of the world as eggs to be broken for one’s own omelet could be the prime directive for an AI bringing about the apocalypse. When Uber wanted more drivers with new cars, its solution was to persuade people with bad credit to take out car loans and then deduct payments directly from their earnings. They positioned this as disrupting the auto loan industry, but everyone else recognized it as predatory lending. The whole idea that disruption is something positive instead of negative is a conceit of tech entrepreneurs. If a superintelligent AI were making a funding pitch to an angel investor, converting the surface of the Earth into strawberry fields would be nothing more than a long overdue disruption of global land use policy.

There are industry observers talking about the need for AIs to have a sense of ethics, and some have proposed that we ensure that any superintelligent AIs we create be “friendly,” meaning that their goals are aligned with human goals. I find these suggestions ironic given that we as a society have failed to teach corporations a sense of ethics, that we did nothing to ensure that Facebook’s and Amazon’s goals were aligned with the public good. But I shouldn’t be surprised; the question of how to create friendly AI is simply more fun to think about than the problem of industry regulation, just as imagining what you’d do during the zombie apocalypse is more fun than thinking about how to mitigate global warming.

There have been some impressive advances in AI recently, like AlphaGo Zero, which became the world’s best Go player in a matter of days purely by playing against itself. But this doesn’t make me worry about the possibility of a superintelligent AI “waking up.” (For one thing, the techniques underlying AlphaGo Zero aren’t useful for tasks in the physical world; we are still a long way from a robot that can walk into your kitchen and cook you some scrambled eggs.) What I’m far more concerned about is the concentration of power in Google, Facebook, and Amazon. They’ve achieved a level of market dominance that is profoundly anticompetitive, but because they operate in a way that doesn’t raise prices for consumers, they don’t meet the traditional criteria for monopolies and so they avoid antitrust scrutiny from the government. We don’t need to worry about Google’s DeepMind research division, we need to worry about the fact that it’s almost impossible to run a business online without using Google’s services.

It’d be tempting to say that fearmongering about superintelligent AI is a deliberate ploy by tech behemoths like Google and Facebook to distract us from what they themselves are doing, which is selling their users’ data to advertisers. If you doubt that’s their goal, ask yourself, why doesn’t Facebook offer a paid version that’s ad free and collects no private information? Most of the apps on your smartphone are available in premium versions that remove the ads; if those developers can manage it, why can’t Facebook? Because Facebook doesn’t want to. Its goal as a company is not to connect you to your friends, it’s to show you ads while making you believe that it’s doing you a favor because the ads are targeted.

So it would make sense if Mark Zuckerberg were issuing the loudest warnings about AI, because pointing to a monster on the horizon would be an effective red herring. But he’s not; he’s actually pretty complacent about AI. The fears of superintelligent AI are probably genuine on the part of the doomsayers. That doesn’t mean they reflect a real threat; what they reflect is the inability of technologists to conceive of moderation as a virtue. Billionaires like Bill Gates and Elon Musk assume that a superintelligent AI will stop at nothing to achieve its goals because that’s the attitude they adopted. (Of course, they saw nothing wrong with this strategy when they were the ones engaging in it; it’s only the possibility that someone else might be better at it than they were that gives them cause for concern.)

Silicon Valley has unconsciously created a devil in their own image, a boogeyman whose excesses are precisely their own.

There’s a saying, popularized by Fredric Jameson, that it’s easier to imagine the end of the world than to imagine the end of capitalism. It’s no surprise that Silicon Valley capitalists don’t want to think about capitalism ending. What’s unexpected is that the way they envision the world ending is through a form of unchecked capitalism, disguised as a superintelligent AI. They have unconsciously created a devil in their own image, a boogeyman whose excesses are precisely their own.

Which brings us back to the importance of insight. Sometimes insight arises spontaneously, but many times it doesn’t. People often get carried away in pursuit of some goal, and they may not realize it until it’s pointed out to them, either by their friends and family or by their therapists. Listening to wake-up calls of this sort is considered a sign of mental health.

We need for the machines to wake up, not in the sense of computers becoming self-aware, but in the sense of corporations recognizing the consequences of their behavior. Just as a superintelligent AI ought to realize that covering the planet in strawberry fields isn’t actually in its or anyone else’s best interests, companies in Silicon Valley need to realize that increasing market share isn’t a good reason to ignore all other considerations. Individuals often reevaluate their priorities after experiencing a personal wake-up call. What we need is for companies to do the same — not to abandon capitalism completely, just to rethink the way they practice it. We need them to behave better than the AIs they fear and demonstrate a capacity for insight.