Efforts to understand the universe could get a boost from an AI developed by Google DeepMind. The algorithm, which can reduce unwanted noise by up to 100 times, could allow the Laser Interferometer Gravitational-Wave Observatory (LIGO) to spot a particular type of black hole that has so far eluded us.

LIGO is designed to detect the gravitational waves produced when objects such as black holes spiral into each other and collide. These waves cross the universe at the speed of light, but the fluctuations they cause in space-time are extremely small 鈥� . Since its first observations 10 years ago, LIGO has recorded such signals produced by nearly 100 black hole collisions.

To do so, the experiment consists of two observatories in the US, each with two arms 4 kilometres long that are set perpendicular to each other. Lasers are beamed down each arm, reflected by precise mirrors at the end and then compared using an interferometer. The length of the arms is changed by a tiny amount as gravitational waves wash over them, and this is carefully recorded to build a picture of the origin of these signals.

The problem is that such demanding accuracy is required that even distant ocean waves or clouds passing overhead can affect measurements. This noise can easily drown out signals, making some observations impossible. Dozens of major adjustments need to be made to filter out the worst of this noise, tweaking the orientation of mirrors and other equipment.

at the California Institute of Technology in Pasadena, who worked with DeepMind to develop the new AI technology, says that attempting to automate these adjustments can ironically create more noise. 鈥淭hat controls noise has been bedevilling us for decades and decades 鈥� everything in this field has been blocked,鈥� says Adhikari. 鈥淗ow do you hold the mirrors so still without inducing noise? If you don鈥檛 control them, the mirrors swing all over the place, and if you control it too much, then it sort of buzzes around.鈥�

at the University of Portsmouth in the UK was one of the scientists who used to manually make these tweaks at LIGO. 鈥淎s you move one thing, something else goes, and something else goes and something else goes,鈥� she says. 鈥淵ou鈥檇 spend forever tweaking.鈥�

Subscriber-only newsletter

Sign up to Lost in Space-Time

Sign up to newsletter

DeepMind鈥檚 new Deep Loop Shaping AI aims to reduce the level of noise from adjusting the mirrors at LIGO by up to 100 times. The AI was trained in a simulation before testing in the real world, and is effectively tasked with achieving two goals: reducing noise and minimising the number of adjustments it makes. 鈥淥ver time, by repeatedly doing it 鈥� it鈥檚 like hundreds and thousands of trials that are running in simulation 鈥� the controller will sort of find what works and what doesn鈥檛 work and find a really, really good policy,鈥� says Jonas Buchli at DeepMind.

at the University of Birmingham, UK, who wasn鈥檛 involved in the research but works on LIGO, says the AI is exciting, although there are many hurdles yet to overcome.

Firstly, the technology has only been run for an hour in the real world on LIGO, so it needs to be shown that it can operate for weeks or even months at a time. Secondly, the technology has so far only been applied to one aspect of control, helping to stabilise the mirrors, and there are hundreds if not thousands of aspects it could conceivably be applied to.

鈥淚t鈥檚 clearly just the first step, but I still think it鈥檚 a very intriguing one. And clearly there is plenty of room for enormous progress,鈥� says Vecchio.

If similar improvements could be made across the board, then he believes we could spot so-called intermediate-size black holes 鈥� for example those with masses around 1000 times that of our sun 鈥� a class of objects without any confirmed observations. The improvements would tend to occur on lower-frequency gravitational waves, where the length of wave is more susceptible to noise, and which are created by larger objects.

鈥淲e know black holes up to 100 solar masses. We know the black holes in our galaxy that are a million solar masses and above. What鈥檚 in between?鈥� says Vecchio. 鈥淧eople think there will be black holes at all these different mass ranges, but nobody has got uncontroversial experimental observational evidence.鈥�

Nuttall says that the new approach could also provide more detailed observation of the types of black hole we have already seen. 鈥淭his is looking pretty damn good,鈥� she says. 鈥淚鈥檓 super excited by this.鈥�

Mysteries of the universe: Cheshire, England

Spend a weekend with some of the brightest minds in science, as you explore the mysteries of the universe in an exciting programme that includes an excursion to see the iconic Lovell Telescope.

Find out more

Google DeepMind claims its latest weather forecasting AI can make predictions faster and more accurately than existing physics-based simulations.

GenCast is the latest in DeepMind鈥檚 ongoing research project to use artificial intelligence to improve weather forecasting. The model was trained on four decades of historical data from the European Centre for Medium-Range Weather Forecasts鈥檚 (ECMWF) , which includes regular measurements of temperature, wind speed and pressure at various altitudes around the globe.

Data up to 2018 was used to train the model and then data from 2019 was used to test its predictions against known weather. The company found that it beat ECMWF鈥檚 industry-standard ENS forecast 97.4 per cent of the time in total, and 99.8 per cent of the time when looking ahead more than 36 hours.

Last year, DeepMind worked with ECMWF to create an AI that beat the 鈥済old-standard鈥� high-resolution HRES 10-day forecast more than 90 per cent of the time. Prior to that, it had developed 鈥渘owcasting鈥� models that predicted the chance of rain in a given 1-square-kilometre area from 5 to 90 minutes ahead using 5 minutes of radar data. And Google is also working on ways of using AI to replace small parts of deterministic models to speed up computation while retaining accuracy.

Existing weather forecasts are based on physics simulations run on powerful supercomputers that deterministically model and extrapolate weather patterns as accurately as possible. Forecasters usually run dozens of simulations with slightly different inputs in groups called ensembles to better capture a range of possible outcomes. These increasingly complex and numerous simulations are extremely computationally intensive and require ever more powerful and energy-hungry machines to operate.

AI could offer a less costly solution. For instance, GenCast creates forecasts with an ensemble of 50 possible futures, each taking just 8 minutes on a custom-made and AI-focused Google Cloud TPU v5 chip.

Free newsletter

Sign up to The Daily

Sign up to newsletter

GenCast operates with a resolution of around 28 square kilometres at the equator. Since the data used in this research was collected, ECMWF鈥檚 ENS has been upgraded to a resolution of just 9 square kilometres.

at DeepMind says the AI may not need to follow suit and could offer a way forward without collecting finer data and running more intensive calculations. 鈥淲hen you have a traditional physics-based model, that is a necessary requirement for getting more accurate predictions, because it鈥檚 a necessary requirement of more accurately solving the physical equations,鈥� says Price. 鈥淸With] machine learning, [it] isn鈥檛 necessarily the case that going to higher resolution is a requirement for getting more accurate simulations or predictions out of your model.鈥�

at the University of Manchester, UK, says AI models present an opportunity to make weather forecasts more efficient but they are often overhyped, and it is important to remember that they rely heavily on training data from traditional physics-based models.

鈥淚s it [GenCast] going to revolutionise numerical weather prediction? No, because you still have to run the numerical weather prediction models in the first place to train the models,鈥� says Schultz. 鈥淚f you never had ECMWF in the first place, creating the ERA5 reanalyses, and all the investment that went into that, you wouldn鈥檛 have these AI tools. That鈥檚 like saying 鈥業 can beat Garry Kasparov at chess, but only after I study every move he ever played鈥�.鈥�

at the US National Oceanic and Atmospheric Administration (NOAA) thinks the AI will need training data with higher resolution to progress further. 鈥淲hat we鈥檙e fundamentally seeing is that all these approaches are getting stopped [from advancing] by the fidelity of training data,鈥� he says. 鈥淎nd the training data comes from operational centres like ECMWF and NOAA. To move this field forward, we need to generate more training data with physics-based models of higher fidelity.鈥�

But for now, GenCast does offer a way to run forecasts at lower computation cost, and more quickly. at the University of Reading, UK, says just as a collection of physics-based forecasts can generate better results than a single forecast, he believes ensembles will boost the accuracy of AI forecasts.

Hunt points to the record 40掳C (104掳F) temperatures seen in the UK in 2022 as an example. A week or two earlier, there were lone members of ensembles predicting it, but they were considered anomalous. Then, as we drew nearer to the heatwave, more and more forecasts fell in line, allowing early warning that something unusual was coming.

鈥淚t does allow you to hedge a little if there is one member that shows something really extreme; it might happen, but it probably won鈥檛,鈥� says Hunt. 鈥淚 wouldn鈥檛 view it as necessarily a step change.听 It鈥檚 combining the tools that we鈥檝e been using in weather forecasting for a while with the new AI approach in a way that will certainly work to improve the quality of AI weather forecasts. I鈥檝e no doubt this will do better than the kind of first wave of AI weather forecasts.鈥�

Google has been using artificial intelligence watermarking to automatically identify text generated by the company鈥檚 Gemini chatbot, making it easier to distinguish AI-generated content from human-written posts. That watermark system could help prevent misuse of the AI chatbots for misinformation and disinformation 鈥� not to mention cheating in school and business settings.

Now, the tech company is making an open-source version of its technique available so that other generative AI developers can similarly watermark the output from their own large language models, says at Google DeepMind, the company鈥檚 AI research team, which combines the former Google Brain and DeepMind labs. 鈥淲hile SynthID isn鈥檛 a silver bullet for identifying AI-generated content, it is an important building block for developing more reliable AI identification tools,鈥� he says.

Independent researchers voiced similar optimism. 鈥淲hile no known watermarking method is foolproof, I really think this can help in catching some fraction of AI-generated misinformation, academic cheating and more,鈥� says at The University of Texas at Austin, who previously worked on AI safety at OpenAI. 鈥淚 hope that other large language model companies, including OpenAI and Anthropic, will follow DeepMind鈥檚 lead on this.鈥�

In May of this year, Google DeepMind that it had implemented its SynthID method for watermarking AI-generated text and video from Google鈥檚 Gemini and Veo AI services, respectively. The company has now published a paper in the journal Nature听showing how SynthID generally outperformed similar AI watermarking techniques for text. The comparison involved assessing how readily responses from various watermarked AI models could be detected.

In Google DeepMind鈥檚 AI watermarking approach, as the model generates a sequence of text, a 鈥渢ournament sampling鈥� algorithm subtly nudges it toward selecting certain word 鈥渢okens鈥�, creating a statistical signature that is detectable by associated software. This process randomly pairs up possible word tokens in a tournament-style bracket, with the winner of each pair being determined by which one scores highest according to a watermarking function. The winners move through successive tournament rounds until just one remains 鈥� a 鈥渕ulti-layered approach鈥� that 鈥渋ncreases the complexity of any potential attempts to reverse-engineer or remove the watermark鈥�, says at the University of Maryland.

A 鈥渄etermined adversary鈥� with huge amounts of computational power could still remove such AI watermarks, says at Harvard University. But he described SynthID鈥檚 approach as making sense given the need for scalable watermarking in AI services.

Free newsletter

Sign up to The Weekly

Sign up to newsletter

The Google DeepMind researchers tested two versions of SynthID that represent trade-offs between making the watermark signature more detectable, at the expense of distorting the text typically generated by an AI model. They showed that the non-distortionary version of the AI watermark still worked, without noticeably affecting the quality of 20 million Gemini-generated text responses during a live experiment.

But the researchers also acknowledged that the watermarking works best with longer chatbot responses that can be answered in a variety of ways 鈥� such as generating an essay or email 鈥� and said it has not yet been tested on responses to maths or coding problems.

Both Google DeepMind鈥檚 team and others described the need for additional safeguards against misuse of AI chatbots 鈥� with Huang recommending stronger regulation as well. 鈥淢andating watermarking by law would address both the practicality and user adoption challenges, ensuring a more secure use of large language models,鈥� she says.

Google DeepMind says its artificial intelligence has helped design chips that are already being used in data centres and even smartphones. But some chip design experts are sceptical of the company鈥檚 claims that such AI can plan new chip layouts better than humans can.

The newly named AlphaChip method can design 鈥渟uperhuman chip layouts鈥� in hours, rather than relying on weeks or months of human effort, said and , researchers at Google DeepMind, in a . This AI approach uses reinforcement learning to figure out the relationships among chip components and gets rewarded based on the final layout quality. But independent researchers say the company has not yet proven such AI can outperform expert human chip designers or commercial software tools 鈥� and they want to see AlphaChip鈥檚 performance on public benchmarks involving current, state-of-the-art circuit designs.

鈥淚f Google would provide experimental results for these designs, we could have fair comparisons, and I expect that everyone would accept the results,鈥� says at Binghamton University in New York. 鈥淭he experiments would take at most a day or two to run, and Google has near-infinite resources 鈥� that these results have not been offered speaks volumes to me.鈥�

Google DeepMind鈥檚 blog post accompanies an to Google鈥檚 2021 Nature journal paper about the company鈥檚 AI process. Since that time, Google DeepMind says that AlphaChip has helped design three generations of Google鈥檚 Tensor Processing Units (TPU) 鈥� specialised chips used to train and run generative AI models for services such as Google鈥檚 Gemini chatbot.

The company also claims that the AI-assisted chip designs perform better than those designed by human experts and have been improving steadily. The AI achieves this by reducing the total length of wires required to connect chip components 鈥� a factor that can lower chip power consumption and potentially improve processing speed. And Google DeepMind says that AlphaChip has created layouts for general-purpose chips used in Google鈥檚 data centres, along with helping the company MediaTek develop a chip used in Samsung mobile phones.

鈥淲e really don鈥檛 know what AlphaChip is today, what it does and what it doesn鈥檛 do,鈥� says , a chip design researcher at a competing firm. 鈥淲e do know that reinforcement learning takes two to three orders of magnitude greater compute resources than methods used in commercial tools and is usually behind [in terms of] results.鈥�

Free newsletter

Sign up to The Weekly

Sign up to newsletter

Markov and Madden critiqued the听original paper鈥檚 claims听about AlphaChip outperforming unnamed human experts. 鈥淐omparisons to unnamed human designers are subjective, not reproducible, and very easy to game. The human designers may be applying low effort or be poorly qualified 鈥� there is no scientific result here,鈥� says Markov. 鈥淚magine if AlphaGo reported wins over unnamed Go players.鈥� A Google DeepMind spokesperson described the experts as members of Google鈥檚 TPU chip design team using the best available commercial tools.

In 2023, an independent expert who had reviewed Google鈥檚 paper his Nature commentary article that had originally praised Google鈥檚 work but had also urged replication. That expert, at the University of California, San Diego, also ran a that tried to replicate Google鈥檚 AI method and found it did not consistently outperform a human expert or conventional computer algorithms. The best-performing methods used for comparison were commercial software or internal research tools for chip design from companies such as Cadence and NVIDIA. In a , Goldie and Mirhoseini disputed Kahng鈥檚 benchmarking results. They said his tests had not pretrained the AI method on specific chip designs 鈥� a crucial factor in its performance 鈥� and relied upon 鈥渇ar fewer compute resources鈥� than Google DeepMind鈥檚 team to train the AI.听

鈥淥n every benchmark where there鈥檚 what I would consider a fair comparison, it seems like reinforcement learning lags behind the state of the art by a wide margin,鈥� says Madden. 鈥淔or circuit placement, I don鈥檛 believe that it鈥檚 a promising research direction.鈥�

An AI-generated recreation of the classic computer game Doom can be played normally despite having no computer code or graphics. Researchers behind the project say similar AI models could be used to create games from scratch in the future, just as they create text and images today.

The model, called , was made by Dani Valevski at Google Research and his colleagues, who declined to speak to New 女生小视频. According to their on the research, the AI can be played for up to 20 seconds while retaining all the features of the original, such as scores, ammunition levels and map layouts. Players can attack enemies, open doors and interact with the environment as usual.

After this period, the model begins to run out of memory and the illusion falls apart.

The original Doom was released in 1993 and has become a popular subject for computer science projects in the years since, including attempts to such as toasters, treadmills and espresso machines.

But in all those cases, the hardware is simply running the original game鈥檚 code. What GameNGen does is fundamentally different: a type of AI called a neural network has learned by observation how to recreate the game without seeing any of its code.

The researchers first created an AI model that learned to interact with Doom as a human would. That model was then tasked with playing the game over and over again while a second AI model, based on the image generator, learned how hundreds of millions of inputs resulted in changes in the game state.

Free newsletter

Sign up to The Daily

Sign up to newsletter

That second model essentially then became a copy of the game, with all of the knowledge, rules and instructions from the original code encoded in the mysterious network of artificial neurons in its own architecture. In tests, human players were only slightly better than random chance at distinguishing short clips of the game from clips of the AI simulation.

GameNGen鈥檚 creators claim in their paper that it is a proof-of-concept for games being created by a neural network rather than lines of code. They suggest that games could be generated from text descriptions or concept art, which would make production less costly than using human programmers.

at the University of Surrey, UK, says the idea of getting a neural network to hallucinate a game environment, and the interactions a human has with it, is an interesting step forward, but not one that will replace human game designers.

鈥淚 don鈥檛 think it鈥檚 the end of those game studios. I think what the game studios have is the imagination, the skills, to actually create these worlds, to understand gameplay, to understand engagement, understand how to draw us into a story. It鈥檚 not just the nuts and bolts, the bits and bytes,鈥� he says. 鈥淭here鈥檚 something very human about creating engaging experiences that we as human beings enjoy that, at the moment, and for the foreseeable future, will largely come from other human beings.鈥�

A new robot hand provides extremely fast and flexible finger movements, while also being tough enough to survive intense damage. That durability helps the hand, which is already being used in Google DeepMind鈥檚 robotics experiments, during the trial-and-error learning required to train artificial intelligence.

This latest robotic hand developed by the UK-based Shadow Robot Company can go from fully open to closed within 500 milliseconds and perform a fingertip pinch with up to 10 newtons of force. It can also withstand repeated punishment such as pistons punching the fingers from multiple angles or a person smashing the device with a hammer.

The new hand鈥檚 robust design is well suited for AI-powered robotics experiments based on reinforcement learning, which allows robots to gradually learn how to interact with environments by fumbling through tasks using trial and error, says at the University of Edinburgh in the UK.

鈥淎ny interaction with the world is a collision damage risk,鈥� said , director of the Shadow Robot Company, during a press conference.

One trade-off is that the hand is 鈥渉eavier than some other options because the design decisions are aimed at reliability over long-term usage鈥�, says Ramamoorthy. The new Shadow Hand鈥檚 chunky, three-fingered set-up weighs 4.1 kilograms in total and 1.2 kilograms per finger.

Free newsletter

Sign up to The Weekly

Sign up to newsletter

This structure makes the hand look much less anthropomorphic than some other robotic limbs, but also makes it more versatile: it can be modified with more fingers if needed, and each finger is a modular component that can be swiftly swapped out for a replacement in case of damage.

Each robotic finger has hundreds of sensors on its fingertips and dozens on the other finger segments. Tiny cameras focus on the inside surface of each robotic finger鈥檚 silicone skin 鈥� touching an object may deform the robotic skin, and this interior view can thus indicate the object鈥檚 hardness and shape.

鈥淚t鈥檚 excellent in terms of sensing, and it鈥檚 super robust,鈥� says at the University of Oxford. 鈥淚t鈥檚 also engineered to be easily fixable.鈥�

Some research labs and companies beyond Google DeepMind may find the capable hand to be useful, says Posner. But it is likely to be expensive 鈥� Shadow Robot Company has not yet announced its price 鈥� and other researchers may prefer cheaper robotic hand options, even if they lack some or all of the sophisticated sensing and object-handling capabilities that the Shadow Hand combines in one package.

Google DeepMind claims to have made the first ever scientific discovery with an AI chatbot by building a fact-checker to filter out useless outputs, leaving only reliable solutions to mathematical or computing problems.

Previous DeepMind achievements, such as using AI to predict the weather or protein shapes, have relied on models created specifically for the task at hand, trained on accurate and specific data. Large language models (LLMs), such as GPT-4 and Google鈥檚 Gemini, are instead trained on vast amounts of varied data to create a breadth of abilities. But that approach also makes them susceptible to 鈥渉allucination鈥�, a term researchers use for producing false outputs.

Gemini 鈥� which was released earlier this month 鈥� has already demonstrated a propensity for hallucination, getting even simple facts such as the . Google鈥檚 previous AI-powered search engine even made errors in the advertising material for its own launch.

One common fix for this phenomenon is to add a layer above the AI that verifies the accuracy of its outputs before passing them to the user. But creating a comprehensive safety net is an enormously difficult task given the broad range of topics that chatbots can be asked about.

at Google DeepMind and his colleagues have created a generalised LLM called FunSearch based on Google鈥檚 PaLM2 model with a fact-checking layer, which they call an 鈥渆valuator鈥�. The model is constrained to providing computer code that solves problems in mathematics and computer science, which DeepMind says is a much more manageable task because these new ideas and solutions are inherently and quickly verifiable.

The underlying AI can still hallucinate and provide inaccurate or misleading results, but the evaluator filters out erroneous outputs and leaves only reliable, potentially useful concepts.

Free newsletter

Sign up to The Daily

Sign up to newsletter

鈥淲e think that perhaps 90 per cent of what the LLM outputs is not going to be useful,鈥� says Fawzi. 鈥淕iven a candidate solution, it鈥檚 very easy for me to tell you whether this is actually a correct solution and to evaluate the solution, but actually coming up with a solution is really hard. And so mathematics and computer science fit particularly well.鈥�

DeepMind claims the model can generate new scientific knowledge and ideas 鈥� something LLMs haven鈥檛 done before.

To start with, FunSearch is given a problem and a very basic solution in source code as an input, then it generates a database of new solutions that are checked by the evaluator for accuracy. The best of the reliable solutions are given back to the LLM as inputs with a prompt asking it to improve on the ideas. DeepMind says the system produces millions of potential solutions, which eventually converge on an efficient result 鈥� sometimes surpassing the best known solution.

For mathematical problems, the model writes computer programs that can find solutions rather than trying to solve the problem directly.

Fawzi and his colleagues challenged FunSearch to find solutions to the cap set problem, which involves determining patterns of points where no three points make a straight line. The problem gets rapidly more computationally intensive as the number of points grows. The AI found a solution consisting of 512 points in eight dimensions, larger than any previously known.

When tasked with the bin-packing problem, where the aim is to efficiently place objects of various sizes into containers, FunSearch found solutions that outperform commonly used algorithms 鈥� a result that has immediate applications for transport and logistics companies. DeepMind says FunSearch could lead to improvements in many more mathematical and computing problems.

at the University of Birmingham, UK, says the next breakthroughs in AI won鈥檛 come from scaling-up LLMs to ever-larger sizes, but from adding layers that ensure accuracy, as DeepMind has done with FunSearch.

鈥淭he strength of a language model is its ability to imagine things, but the problem is hallucinations,鈥� says Lee. 鈥淎nd this research is breaking that problem: it鈥檚 reining it in, or fact-checking. It鈥檚 a neat idea.鈥�

Lee says AIs shouldn鈥檛 be criticised for producing large amounts of inaccurate or useless outputs, as this is not dissimilar to the way that human mathematicians and scientists operate: brainstorming ideas, testing them and following up on the best ones while discarding the worst.