Health care
Group items matching
in title, tags, annotations or urlTwo years to save the world, says Sunak's AI adviser - 0 views
AI is about to completely change how you use computers | Bill Gates - 0 views
-
Today, AI’s main role in healthcare is to help with administrative tasks. Abridge, Nuance DAX, and Nabla Copilot, for example, can capture audio during an appointment and then write up notes for the doctor to review.
- ...38 more annotations...
-
Already, AI can help you pick out a new TV and recommend movies, books, shows, and podcasts. Likewise, a company I’ve invested in, recently launched Pix, which lets you ask questions (“Which Robert Redford movies would I like and where can I watch them?”) and then makes recommendations based on what you’ve liked in the past
-
copilots can do a lot—such as turn a written document into a slide deck, answer questions about a spreadsheet using natural language, and summarize email threads while representing each person’s point of view.
-
before the sophisticated agents I’m describing become a reality, we need to confront a number of questions about the technology and how we’ll use it.
-
Helping patients and healthcare workers will be especially beneficial for people in poor countries, where many never get to see a doctor at all.
-
To create a new app or service, you won’t need to know how to write code or do graphic design. You’ll just tell your agent what you want. It will be able to write the code, design the look and feel of the app, create a logo, and publish the app to an online store
-
Agents will do even more. Having one will be like having a person dedicated to helping you with various tasks and doing them independently if you want. If you have an idea for a business, an agent will help you write up a business plan, create a presentation for it, and even generate images of what your product might look like
-
For decades, I’ve been excited about all the ways that software would make teachers’ jobs easier and help students learn. It won’t replace teachers, but it will supplement their work—personalizing the work for students and liberating teachers from paperwork and other tasks so they can spend more time on the most important parts of the job.
-
Mental health care is another example of a service that agents will make available to virtually everyone. Today, weekly therapy sessions seem like a luxury. But there is a lot of unmet need, and many people who could benefit from therapy don’t have access to it.
-
I don’t think any single company will dominate the agents business--there will be many different AI engines available.
-
The real shift will come when agents can help patients do basic triage, get advice about how to deal with health problems, and decide whether they need to seek treatment.
-
For example, few families can pay for a tutor who works one-on-one with a student to supplement their classroom work. If agents can capture what makes a tutor effective, they’ll unlock this supplemental instruction for everyone who wants it. If a tutoring agent knows that a kid likes Minecraft and Taylor Swift, it will use Minecraft to teach them about calculating the volume and area of shapes, and Taylor’s lyrics to teach them about storytelling and rhyme schemes. The experience will be far richer—with graphics and sound, for example—and more personalized than today’s text-based tutors.
-
your agent will be able to help you in the same way that personal assistants support executives today. If your friend just had surgery, your agent will offer to send flowers and be able to order them for you. If you tell it you’d like to catch up with your old college roommate, it will work with their agent to find a time to get together, and just before you arrive, it will remind you that their oldest child just started college at the local university.
-
To see the dramatic change that agents will bring, let’s compare them to the AI tools available today. Most of these are bots. They’re limited to one app and generally only step in when you write a particular word or ask for help. Because they don’t remember how you use them from one time to the next, they don’t get better or learn any of your preferences.
-
The current state of the art is Khanmigo, a text-based bot created by Khan Academy. It can tutor students in math, science, and the humanities—for example, it can explain the quadratic formula and create math problems to practice on. It can also help teachers do things like write lesson plans.
-
Businesses that are separate today—search advertising, social networking with advertising, shopping, productivity software—will become one business.
-
other issues won’t be decided by companies and governments. For example, agents could affect how we interact with friends and family. Today, you can show someone that you care about them by remembering details about their life—say, their birthday. But when they know your agent likely reminded you about it and took care of sending flowers, will it be as meaningful for them?
-
In the computing industry, we talk about platforms—the technologies that apps and services are built on. Android, iOS, and Windows are all platforms. Agents will be the next platform.
-
Agents won’t simply make recommendations; they’ll help you act on them. If you want to buy a camera, you’ll have your agent read all the reviews for you, summarize them, make a recommendation, and place an order for it once you’ve made a decision.
-
Agents will affect how we use software as well as how it’s written. They’ll replace search sites because they’ll be better at finding information and summarizing it for you
-
they’ll be dramatically better. You’ll be able to have nuanced conversations with them. They will be much more personalized, and they won’t be limited to relatively simple tasks like writing a letter.
-
Companies will be able to make agents available for their employees to consult directly and be part of every meeting so they can answer questions.
-
AI agents that are well trained in mental health will make therapy much more affordable and easier to get. Wysa and Youper are two of the early chatbots here. But agents will go much deeper. If you choose to share enough information with a mental health agent, it will understand your life history and your relationships. It’ll be available when you need it, and it will never get impatient. It could even, with your permission, monitor your physical responses to therapy through your smart watch—like if your heart starts to race when you’re talking about a problem with your boss—and suggest when you should see a human therapist.
-
If the number of companies that have started working on AI just this year is any indication, there will be an exceptional amount of competition, which will make agents very inexpensive.
-
Agents are smarter. They’re proactive—capable of making suggestions before you ask for them. They accomplish tasks across applications. They improve over time because they remember your activities and recognize intent and patterns in your behavior. Based on this information, they offer to provide what they think you need, although you will always make the final decisions.
-
Agents are not only going to change how everyone interacts with computers. They’re also going to upend the software industry, bringing about the biggest revolution in computing since we went from typing commands to tapping on icons.
-
The most exciting impact of AI agents is the way they will democratize services that today are too expensive for most people
-
In the next five years, this will change completely. You won’t have to use different apps for different tasks. You’ll simply tell your device, in everyday language, what you want to do. And depending on how much information you choose to share with it, the software will be able to respond personally because it will have a rich understanding of your life. In the near future, anyone who’s online will be able to have a personal assistant powered by artificial intelligence that’s far beyond today’s technology.
-
You’ll also be able to get news and entertainment that’s been tailored to your interests. CurioAI, which creates a custom podcast on any subject you ask about, is a glimpse of what’s coming.
-
An agent will be able to help you with all your activities if you want it to. With permission to follow your online interactions and real-world locations, it will develop a powerful understanding of the people, places, and activities you engage in. It will get your personal and work relationships, hobbies, preferences, and schedule. You’ll choose how and when it steps in to help with something or ask you to make a decision.
-
even the best sites have an incomplete understanding of your work, personal life, interests, and relationships and a limited ability to use this information to do things for you. That’s the kind of thing that is only possible today with another human being, like a close friend or personal assistant.
-
In the distant future, agents may even force humans to face profound questions about purpose. Imagine that agents become so good that everyone can have a high quality of life without working nearly as much. In a future like that, what would people do with their time? Would anyone still want to get an education when an agent has all the answers? Can you have a safe and thriving society when most people have a lot of free time on their hands?
-
They’ll have an especially big influence in four areas: health care, education, productivity, and entertainment and shopping.
Instagram's Algorithm Delivers Toxic Video Mix to Adults Who Follow Children - WSJ - 0 views
www.wsj.com/...-adult-sexual-content-72874155
social media research psychology child abuse algorithm sexual
![](/images/link.gif)
-
Instagram’s Reels video service is designed to show users streams of short videos on topics the system decides will interest them, such as sports, fashion or humor.
-
The Meta Platforms META -1.04%decrease; red down pointing triangle-owned social app does the same thing for users its algorithm decides might have a prurient interest in children, testing by The Wall Street Journal showed.
-
The Journal sought to determine what Instagram’s Reels algorithm would recommend to test accounts set up to follow only young gymnasts, cheerleaders and other teen and preteen influencers active on the platform.
- ...30 more annotations...
-
“Our systems are effective at reducing harmful content, and we’ve invested billions in safety, security and brand suitability solutions,” said Samantha Stetson, a Meta vice president who handles relations with the advertising industry. She said the prevalence of inappropriate content on Instagram is low, and that the company invests heavily in reducing it.
-
The Journal set up the test accounts after observing that the thousands of followers of such young people’s accounts often include large numbers of adult men, and that many of the accounts who followed those children also had demonstrated interest in sex content related to both children and adults
-
The Journal also tested what the algorithm would recommend after its accounts followed some of those users as well, which produced more-disturbing content interspersed with ads.
-
The Canadian Centre for Child Protection, a child-protection group, separately ran similar tests on its own, with similar results.
-
Meta said the Journal’s tests produced a manufactured experience that doesn’t represent what billions of users see. The company declined to comment on why the algorithms compiled streams of separate videos showing children, sex and advertisements, but a spokesman said that in October it introduced new brand safety tools that give advertisers greater control over where their ads appear, and that Instagram either removes or reduces the prominence of four million videos suspected of violating its standards each month.
-
The Journal reported in June that algorithms run by Meta, which owns both Facebook and Instagram, connect large communities of users interested in pedophilic content. The Meta spokesman said a task force set up after the Journal’s article has expanded its automated systems for detecting users who behave suspiciously, taking down tens of thousands of such accounts each month. The company also is participating in a new industry coalition to share signs of potential child exploitation.
-
Following what it described as Meta’s unsatisfactory response to its complaints, Match began canceling Meta advertising for some of its apps, such as Tinder, in October. It has since halted all Reels advertising and stopped promoting its major brands on any of Meta’s platforms. “We have no desire to pay Meta to market our brands to predators or place our ads anywhere near this content,” said Match spokeswoman Justine Sacco.
-
Even before the 2020 launch of Reels, Meta employees understood that the product posed safety concerns, according to former employees.
-
Robbie McKay, a spokesman for Bumble, said it “would never intentionally advertise adjacent to inappropriate content,” and that the company is suspending its ads across Meta’s platforms.
-
Meta created Reels to compete with TikTok, the video-sharing platform owned by Beijing-based ByteDance. Both products feed users a nonstop succession of videos posted by others, and make money by inserting ads among them. Both companies’ algorithms show to a user videos the platforms calculate are most likely to keep that user engaged, based on his or her past viewing behavior
-
The Journal reporters set up the Instagram test accounts as adults on newly purchased devices and followed the gymnasts, cheerleaders and other young influencers. The tests showed that following only the young girls triggered Instagram to begin serving videos from accounts promoting adult sex content alongside ads for major consumer brands, such as one for Walmart that ran after a video of a woman exposing her crotch.
-
When the test accounts then followed some users who followed those same young people’s accounts, they yielded even more disturbing recommendations. The platform served a mix of adult pornography and child-sexualizing material, such as a video of a clothed girl caressing her torso and another of a child pantomiming a sex act.
-
Experts on algorithmic recommendation systems said the Journal’s tests showed that while gymnastics might appear to be an innocuous topic, Meta’s behavioral tracking has discerned that some Instagram users following preteen girls will want to engage with videos sexualizing children, and then directs such content toward them.
-
Current and former Meta employees said in interviews that the tendency of Instagram algorithms to aggregate child sexualization content from across its platform was known internally to be a problem. Once Instagram pigeonholes a user as interested in any particular subject matter, they said, its recommendation systems are trained to push more related content to them.
-
Preventing the system from pushing noxious content to users interested in it, they said, requires significant changes to the recommendation algorithms that also drive engagement for normal users. Company documents reviewed by the Journal show that the company’s safety staffers are broadly barred from making changes to the platform that might reduce daily active users by any measurable amount.
-
The test accounts showed that advertisements were regularly added to the problematic Reels streams. Ads encouraging users to visit Disneyland for the holidays ran next to a video of an adult acting out having sex with her father, and another of a young woman in lingerie with fake blood dripping from her mouth. An ad for Hims ran shortly after a video depicting an apparently anguished woman in a sexual situation along with a link to what was described as “the full video.”
-
Instagram’s system served jarring doses of salacious content to those test accounts, including risqué footage of children as well as overtly sexual adult videos—and ads for some of the biggest U.S. brands.
-
Part of the problem is that automated enforcement systems have a harder time parsing video content than text or still images. Another difficulty arises from how Reels works: Rather than showing content shared by users’ friends, the way other parts of Instagram and Facebook often do, Reels promotes videos from sources they don’t follow
-
In an analysis conducted shortly before the introduction of Reels, Meta’s safety staff flagged the risk that the product would chain together videos of children and inappropriate content, according to two former staffers. Vaishnavi J, Meta’s former head of youth policy, described the safety review’s recommendation as: “Either we ramp up our content detection capabilities, or we don’t recommend any minor content,” meaning any videos of children.
-
At the time, TikTok was growing rapidly, drawing the attention of Instagram’s young users and the advertisers targeting them. Meta didn’t adopt either of the safety analysis’s recommendations at that time, according to J.
-
Stetson, Meta’s liaison with digital-ad buyers, disputed that Meta had neglected child safety concerns ahead of the product’s launch. “We tested Reels for nearly a year before releasing it widely, with a robust set of safety controls and measures,” she said.
-
After initially struggling to maximize the revenue potential of its Reels product, Meta has improved how its algorithms recommend content and personalize video streams for users
-
Among the ads that appeared regularly in the Journal’s test accounts were those for “dating” apps and livestreaming platforms featuring adult nudity, massage parlors offering “happy endings” and artificial-intelligence chatbots built for cybersex. Meta’s rules are supposed to prohibit such ads.
-
The Journal informed Meta in August about the results of its testing. In the months since then, tests by both the Journal and the Canadian Centre for Child Protection show that the platform continued to serve up a series of videos featuring young children, adult content and apparent promotions for child sex material hosted elsewhere.
-
As of mid-November, the center said Instagram is continuing to steadily recommend what the nonprofit described as “adults and children doing sexual posing.”
-
Meta hasn’t offered a timetable for resolving the problem or explained how in the future it would restrict the promotion of inappropriate content featuring children.
-
The Journal’s test accounts found that the problem even affected Meta-related brands. Ads for the company’s WhatsApp encrypted chat service and Meta’s Ray-Ban Stories glasses appeared next to adult pornography. An ad for Lean In Girls, the young women’s empowerment nonprofit run by former Meta Chief Operating Officer Sheryl Sandberg, ran directly before a promotion for an adult sex-content creator who often appears in schoolgirl attire. Sandberg declined to comment.
-
Through its own tests, the Canadian Centre for Child Protection concluded that Instagram was regularly serving videos and pictures of clothed children who also appear in the National Center for Missing and Exploited Children’s digital database of images and videos confirmed to be child abuse sexual material. The group said child abusers often use the images of the girls to advertise illegal content for sale in dark-web forums.
-
The nature of the content—sexualizing children without generally showing nudity—reflects the way that social media has changed online child sexual abuse, said Lianna McDonald, executive director for the Canadian center. The group has raised concerns about the ability of Meta’s algorithms to essentially recruit new members of online communities devoted to child sexual abuse, where links to illicit content in more private forums proliferate.
-
“Time and time again, we’ve seen recommendation algorithms drive users to discover and then spiral inside of these online child exploitation communities,” McDonald said, calling it disturbing that ads from major companies were subsidizing that process.
Opinion | The Imminent Danger of A.I. Is One We're Not Talking About - The New York Times - 0 views
-
a void at the center of our ongoing reckoning with A.I. We are so stuck on asking what the technology can do that we are missing the more important questions: How will it be used? And who will decide?
-
“Sydney” is a predictive text system built to respond to human requests. Roose wanted Sydney to get weird — “what is your shadow self like?” he asked — and Sydney knew what weird territory for an A.I. system sounds like, because human beings have written countless stories imagining it. At some point the system predicted that what Roose wanted was basically a “Black Mirror” episode, and that, it seems, is what it gave him. You can see that as Bing going rogue or as Sydney understanding Roose perfectly.
- ...22 more annotations...
-
The question at the core of the Roose/Sydney chat is: Who did Bing serve? We assume it should be aligned to the interests of its owner and master, Microsoft. It’s supposed to be a good chatbot that politely answers questions and makes Microsoft piles of money. But it was in conversation with Kevin Roose. And Roose was trying to get the system to say something interesting so he’d have a good story. It did that, and then some. That embarrassed Microsoft. Bad Bing! But perhaps — good Sydney?
-
Microsoft — and Google and Meta and everyone else rushing these systems to market — hold the keys to the code. They will, eventually, patch the system so it serves their interests. Sydney giving Roose exactly what he asked for was a bug that will soon be fixed. Same goes for Bing giving Microsoft anything other than what it wants.
-
These systems, she said, are terribly suited to being integrated into search engines. “They’re not trained to predict facts,” she told me. “They’re essentially trained to make up things that look like facts.”
-
That’s where things get scary. Roose described Sydney’s personality as “very persuasive and borderline manipulative.” It was a striking comment
-
this technology will become what it needs to become to make money for the companies behind it, perhaps at the expense of its users.
-
What about when these systems are deployed on behalf of the scams that have always populated the internet? How about on behalf of political campaigns? Foreign governments? “I think we wind up very fast in a world where we just don’t know what to trust anymore,”
-
What if they worked much, much better? What if Google and Microsoft and Meta and everyone else end up unleashing A.I.s that compete with one another to be the best at persuading users to want what the advertisers are trying to sell?
-
Large language models, as they’re called, are built to persuade. They have been trained to convince humans that they are something close to human. They have been programmed to hold conversations, responding with emotion and emoji
-
They are being turned into friends for the lonely and assistants for the harried. They are being pitched as capable of replacing the work of scores of writers and graphic designers and form-fillers
-
They are the ones who have anthropomorphized these systems, making them sound like humans rather than keeping them recognizably alien.
-
I’d feel better, for instance, about an A.I. helper I paid a monthly fee to use rather than one that appeared to be free
-
It’s possible, for example, that the advertising-based models could gather so much more data to train the systems that they’d have an innate advantage over the subscription models
-
Much of the work of the modern state is applying the values of society to the workings of markets, so that the latter serve, to some rough extent, the former
-
We have done this extremely well in some markets — think of how few airplanes crash, and how free of contamination most food is — and catastrophically poorly in others.
-
One danger here is that a political system that knows itself to be technologically ignorant will be cowed into taking too much of a wait-and-see approach to A.I.
-
wait long enough and the winners of the A.I. gold rush will have the capital and user base to resist any real attempt at regulation
-
Somehow, society is going to have to figure out what it’s comfortable having A.I. doing, and what A.I. should not be permitted to try, before it is too late to make those decisions.
-
Most fears about capitalism are best understood as fears about our inability to regulate capitalism.
GPT-4 has arrived. It will blow ChatGPT out of the water. - The Washington Post - 0 views
-
GPT-4, in contrast, is a state-of-the-art system capable of creating not just words but describing images in response to a person’s simple written commands.
-
When shown a photo of a boxing glove hanging over a wooden seesaw with a ball on one side, for instance, a person can ask what will happen if the glove drops, and GPT-4 will respond that it would hit the seesaw and cause the ball to fly up.
-
an AI program, known as a large language model, that early testers had claimed was remarkably advanced in its ability to reason and learn new things
- ...22 more annotations...
-
hose promises have also fueled anxiety over how people will be able to compete for jobs outsourced to eerily refined machines or trust the accuracy of what they see online.
-
Officials with the San Francisco lab said GPT-4’s “multimodal” training across text and images would allow it to escape the chat box and more fully emulate a world of color and imagery, surpassing ChatGPT in its “advanced reasoning capabilities.”
-
A person could upload an image and GPT-4 could caption it for them, describing the objects and scene.
-
AI language models often confidently offer wrong answers because they are designed to spit out cogent phrases, not actual facts. And because they have been trained on internet text and imagery, they have also learned to emulate human biases of race, gender, religion and class.
-
GPT-4 still makes many of the errors of previous versions, including “hallucinating” nonsense, perpetuating social biases and offering bad advice. It also lacks knowledge of events that happened after about September 2021, when its training data was finalized, and “does not learn from its experience,” limiting people’s ability to teach it new things.
-
Microsoft has invested billions of dollars in OpenAI in the hope its technology will become a secret weapon for its workplace software, search engine and other online ambitions. It has marketed the technology as a super-efficient companion that can handle mindless work and free people for creative pursuits, helping one software developer to do the work of an entire team or allowing a mom-and-pop shop to design a professional advertising campaign without outside help.
-
sparked criticism that the companies are rushing to exploit an untested, unregulated and unpredictable technology that could deceive people, undermine artists’ work and lead to real-world harm.
-
the company held back the feature to better understand potential risks. As one example, she said, the model might be able to look at an image of a big group of people and offer up known information about them, including their identities — a possible facial recognition use case that could be used for mass surveillance.
-
OpenAI researchers wrote, “As GPT-4 and AI systems like it are adopted more widely,” they “will have even greater potential to reinforce entire ideologies, worldviews, truths and untruths, and to cement them or lock them in.”
-
“We can agree as a society broadly on some harms that a model should not contribute to,” such as building a nuclear bomb or generating child sexual abuse material, she said. “But many harms are nuanced and primarily affect marginalized groups,” she added, and those harmful biases, especially across other languages, “cannot be a secondary consideration in performance.”
-
OpenAI said its new model would be able to handle more than 25,000 words of text, a leap forward that could facilitate longer conversations and allow for the searching and analysis of long documents.
-
OpenAI developers said GPT-4 was more likely to provide factual responses and less likely to refuse harmless requests
-
Duolingo, the language learning app, has already used GPT-4 to introduce new features, such as an AI conversation partner and a tool that tells users why an answer was incorrect.
-
The company did not share evaluations around bias that have become increasingly common after pressure from AI ethicists.
-
GPT-4 will have competition in the growing field of multisensory AI. DeepMind, an AI firm owned by Google’s parent company Alphabet, last year released a “generalist” model named Gato that can describe images and play video games. And Google this month released a multimodal system, PaLM-E, that folded AI vision and language expertise into a one-armed robot on wheels: If someone told it to go fetch some chips, for instance, it could comprehend the request, wheel over to a drawer and choose the right bag.
-
The systems, though — as critics and the AI researchers are quick to point out — are merely repeating patterns and associations found in their training data without a clear understanding of what it’s saying or when it’s wrong.
-
GPT-4, the fourth “generative pre-trained transformer” since OpenAI’s first release in 2018, relies on a breakthrough neural-network technique in 2017 known as the transformer that rapidly advanced how AI systems can analyze patterns in human speech and imagery.
-
The systems are “pre-trained” by analyzing trillions of words and images taken from across the internet: news articles, restaurant reviews and message-board arguments; memes, family photos and works of art.
-
Giant supercomputer clusters of graphics processing chips are mapped out their statistical patterns — learning which words tended to follow each other in phrases, for instance — so that the AI can mimic those patterns, automatically crafting long passages of text or detailed images, one word or pixel at a time.
-
In 2019, the company refused to publicly release GPT-2, saying it was so good they were concerned about the “malicious applications” of its use, from automated spam avalanches to mass impersonation and disinformation campaigns.
-
Altman has also marketed OpenAI’s vision with the aura of science fiction come to life. In a blog post last month, he said the company was planning for ways to ensure that “all of humanity” benefits from “artificial general intelligence,” or AGI — an industry term for the still-fantastical idea of an AI superintelligence that is generally as smart as, or smarter than, the humans themselves.
AlphaProof, a New A.I. from Google DeepMind, Scores Big at the International Math Olympiad - The New York Times - 0 views
-
Last week the DeepMind researchers got out the gong again to celebrate what Alex Davies, a lead of Google DeepMind’s mathematics initiative, described as a “massive breakthrough” in mathematical reasoning by an A.I. system.
-
A pair of Google DeepMind models tried their luck with the problem set in the 2024 International Mathematical Olympiad, or I.M.O., held from July 11 to July 22 about 100 miles west of London at the University of Bath.
-
The event is said to be the premier math competition for the world’s “brightest mathletes,” according to a promotional post on social media.
- ...17 more annotations...
-
The human problem-solvers — 609 high school students from 108 countries — won 58 gold medals, 123 silver and 145 bronze. The A.I. performed at the level of a silver medalist, solving four out of six problems for a total of 28 points. It was the first time that A.I. has achieved a medal-worthy performance on an Olympiad’s problems.
-
Nonetheless, Dr. Kohli described the result as a “phase transition” — a transformative change — “in the use of A.I. in mathematics and the ability of A.I. systems to do mathematics.”
-
Dr. Gowers added in an email: “I was definitely impressed.” The lab had discussed its Olympiad ambitions with him a couple of weeks beforehand, so “my expectations were quite high,” he said. “But the program met them, and in one or two instances significantly surpassed them.” The program found the “magic keys” that unlocked the problems, he said.
-
Haojia Shi, a student from China, ranked No. 1 and was the only competitor to earn a perfect score — 42 points for six problems; each problem is worth seven points for a full solution. The U.S. team won first place with 192 points; China placed second with 190.
-
The Google system earned its 28 points for fully solving four problems — two in algebra, one in geometry and one in number theory. (It flopped at two combinatorics problems.) The system was allowed unlimited time; for some problems it took up to three days. The students were allotted only 4.5 hours per exam.
-
“The fact that we’ve reached this threshold, where it’s even possible to tackle these problems at all, is what represents a step-change in the history of mathematics,” he added. “And hopefully it’s not just a step-change in the I.M.O., but also represents the point at which we went from computers only being able to prove very, very simple things toward computers being able to prove things that humans can’t.”
-
“Mathematics requires this interesting combination of abstract, precise and creative reasoning,” Dr. Davies said. In part, he noted, this repertoire of abilities is what makes math a good litmus test for the ultimate goal: reaching so-called artificial general intelligence, or A.G.I., a system with capabilities ranging from emerging to competent to virtuoso to superhuman
-
In January, a Google DeepMind system named AlphaGeometry solved a sampling of Olympiad geometry problems at nearly the level of a human gold medalist. “AlphaGeometry 2 has now surpassed the gold medalists in solving I.M.O. problems,” Thang Luong, the principal investigator, said in an email.
-
Dr. Hubert’s team developed a new model that is comparable but more generalized. Named AlphaProof, it is designed to engage with a broad range of mathematical subjects. All told, AlphaGeometry and AlphaProof made use of a number of different A.I. technologies.
-
One approach was an informal reasoning system, expressed in natural language. This system leveraged Gemini, Google’s large language model. It used the English corpus of published problems and proofs and the like as training data.
-
The informal system excels at identifying patterns and suggesting what comes next; it is creative and talks about ideas in an understandable way. Of course, large language models are inclined to make things up — which may (or may not) fly for poetry and definitely not for math. But in this context, the L.L.M. seems to have displayed restraint; it wasn’t immune to hallucination, but the frequency was reduced.
-
Another approach was a formal reasoning system, based on logic and expressed in code. It used theorem prover and proof-assistant software called Lean, which guarantees that if the system says a proof is correct, then it is indeed correct. “We can exactly check that the proof is correct or not,” Dr. Hubert said. “Every step is guaranteed to be logically sound.”
-
Another crucial component was a reinforcement learning algorithm in the AlphaGo and AlphaZero lineage. This type of A.I. learns by itself and can scale indefinitely, said Dr. Silver, who is Google DeepMind’s vice-president of reinforcement learning. Since the algorithm doesn’t require a human teacher, it can “learn and keep learning and keep learning until ultimately it can solve the hardest problems that humans can solve,” he said. “And then maybe even one day go beyond those.”
-
Dr. Hubert added, “The system can rediscover knowledge for itself.” That’s what happened with AlphaZero: It started with zero knowledge, Dr. Hubert said, “and by just playing games, and seeing who wins and who loses, it could rediscover all the knowledge of chess. It took us less than a day to rediscover all the knowledge of chess, and about a week to rediscover all the knowledge of Go. So we thought, Let’s apply this to mathematics.”
-
Dr. Gowers doesn’t worry — too much — about the long-term consequences. “It is possible to imagine a state of affairs where mathematicians are basically left with nothing to do,” he said. “That would be the case if computers became better, and far faster, at everything that mathematicians currently do.”
-
“There still seems to be quite a long way to go before computers will be able to do research-level mathematics,” he added. “It’s a fairly safe bet that if Google DeepMind can solve at least some hard I.M.O. problems, then a useful research tool can’t be all that far away.”
-
A really adept tool might make mathematics accessible to more people, speed up the research process, nudge mathematicians outside the box. Eventually it might even pose novel ideas that resonate.
‹ Previous
21 - 27 of 27